releases/4.19.31/mm-vmalloc-fix-size-check-for-remap_vmalloc_range_partial.patch

   1 From 401592d2e095947344e10ec0623adbcd58934dd4 Mon Sep 17 00:00:00 2001
   2 From: Roman Penyaev <rpenyaev@suse.de>
   3 Date: Tue, 5 Mar 2019 15:43:20 -0800
   4 Subject: mm/vmalloc: fix size check for remap_vmalloc_range_partial()
   5
   6 From: Roman Penyaev <rpenyaev@suse.de>
   7
   8 commit 401592d2e095947344e10ec0623adbcd58934dd4 upstream.
   9
  10 When VM_NO_GUARD is not set area->size includes adjacent guard page,
  11 thus for correct size checking get_vm_area_size() should be used, but
  12 not area->size.
  13
  14 This fixes possible kernel oops when userspace tries to mmap an area on
  15 1 page bigger than was allocated by vmalloc_user() call: the size check
  16 inside remap_vmalloc_range_partial() accounts non-existing guard page
  17 also, so check successfully passes but vmalloc_to_page() returns NULL
  18 (guard page does not physically exist).
  19
  20 The following code pattern example should trigger an oops:
  21
  22   static int oops_mmap(struct file *file, struct vm_area_struct *vma)
  23   {
  24         void *mem;
  25
  26         mem = vmalloc_user(4096);
  27         BUG_ON(!mem);
  28         /* Do not care about mem leak */
  29
  30         return remap_vmalloc_range(vma, mem, 0);
  31   }
  32
  33 And userspace simply mmaps size + PAGE_SIZE:
  34
  35   mmap(NULL, 8192, PROT_WRITE|PROT_READ, MAP_PRIVATE, fd, 0);
  36
  37 Possible candidates for oops which do not have any explicit size
  38 checks:
  39
  40    *** drivers/media/usb/stkwebcam/stk-webcam.c:
  41    v4l_stk_mmap[789]   ret = remap_vmalloc_range(vma, sbuf->buffer, 0);
  42
  43 Or the following one:
  44
  45    *** drivers/video/fbdev/core/fbmem.c
  46    static int
  47    fb_mmap(struct file *file, struct vm_area_struct * vma)
  48         ...
  49         res = fb->fb_mmap(info, vma);
  50
  51 Where fb_mmap callback calls remap_vmalloc_range() directly without any
  52 explicit checks:
  53
  54    *** drivers/video/fbdev/vfb.c
  55    static int vfb_mmap(struct fb_info *info,
  56              struct vm_area_struct *vma)
  57    {
  58        return remap_vmalloc_range(vma, (void *)info->fix.smem_start, vma->vm_pgoff);
  59    }
  60
  61 Link: http://lkml.kernel.org/r/20190103145954.16942-2-rpenyaev@suse.de
  62 Signed-off-by: Roman Penyaev <rpenyaev@suse.de>
  63 Acked-by: Michal Hocko <mhocko@suse.com>
  64 Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
  65 Cc: Joe Perches <joe@perches.com>
  66 Cc: "Luis R. Rodriguez" <mcgrof@kernel.org>
  67 Cc: <stable@vger.kernel.org>
  68 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  69 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
  70 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  71
  72 ---
  73  mm/vmalloc.c |    2 +-
  74  1 file changed, 1 insertion(+), 1 deletion(-)
  75
  76 --- a/mm/vmalloc.c
  77 +++ b/mm/vmalloc.c
  78 @@ -2244,7 +2244,7 @@ int remap_vmalloc_range_partial(struct v
  79         if (!(area->flags & VM_USERMAP))
  80                 return -EINVAL;
  81
  82 -       if (kaddr + size > area->addr + area->size)
  83 +       if (kaddr + size > area->addr + get_vm_area_size(area))
  84                 return -EINVAL;
  85
  86         do {